Path Problem in edit_inference.py
Hello, I have to ask you for help.
An error occurred while I was running the reasoning process:
fs2_model_path = "/home/LiuruiGrp/2022xjt/gitcodepro/a3t/egs2/vctk/sedit/exp/conformer/train.loss.ave_5best.pth"
fs2_model, fs2_processor = get_fs2_model(fs2_model_path)
At first, "AttributeError: 'Namespace'object has no attribute 'tts' ", So I manually added the 'tts' and 'tts_conf' information to the 'config.yaml' file, but an error message "TypeError: init() got an unexpected keyword argument 'lsm_weight' " was reported. I still haven't found the solution so far, so I need your help, thank you!
How can I make the file read the 'lsm_weight' configuration information?
Looking forward to your reply.
Best Regards!
Hi,
Can you post the full error message for me to check which line got this error?
Sorry for the long delay in replying you, I'll send you a picture.
Hi
@XiJ
,
fs2_model_path is the path of the fastspeech2 model. but you provided it with an a3t checkpoint.
Hi, I have solved the above problems, and sedit_inference.py was also successfully run, but the synthesized audio effect was extremely poor, and the newly added "Take a look at these pages for crooked creek drive" could not be heard clearly.
Could you tell me the python version used to load the pickle file since i'm getting error "_pickle.UnpicklingError: invalid load key, 'v'."
Could you tell me the python version used to load the pickle file since i'm getting error "_pickle.UnpicklingError: invalid load key, 'v'."
Can you try with this docker? docker pull richardbaihe/pytorch:a3t
And inside this docker, use conda activate espnet
Hi, I have solved the above problems, and sedit_inference.py was also successfully run, but the synthesized audio effect was extremely poor, and the newly added "Take a look at these pages for crooked creek drive" could not be heard clearly.
It should work if you load the checkpoint successfully, since the generations have been reproduced by other github users. Please check if you have loaded the correct checkpoint.